Muhammad Umer S.

I build production Snowflake, Microsoft Fabric, and Databricks data pipelines for teams whose dashboards stall, break silently, or feed decisions on stale numbers. If your BI team is validating the same revenue number against three different reports before a Monday meeting, that isn't a dashboard problem. It's a data model that was never given a single source of truth. I specialize in replacing fragmented, source-by-source reporting logic with governed Snowflake and dbt or Microsoft Fabric medallion architectures, orchestrated end to end with Apache Airflow, so every dashboard pulls from one tested SQL model instead of rebuilding its own calculations. Most data engineers hand off a pipeline that works on day one and go quiet. I build for the day a source changes schema, a job fails at 2am, or a stakeholder asks why last month's numbers shifted. This is production data engineering, not one-off scripts. Incremental processing, CDC based ingestion, automated data quality tests, and traceable lineage are built in from the start, not bolted on after something breaks. Case Study 1: Snowflake + dbt Production ELT Platform - Built a layered ELT platform (staging, intermediate, marts) with incremental processing, CDC based schema drift handling, and dbt tests gating every model before it reached BI, all scheduled and monitored through Apache Airflow. Result: cut manual data reconciliation time by an estimated 8 to 10 hours a week and reduced pipeline failure incidents by roughly 60 to 70%, since every model now fails loudly and traceably instead of silently. Case Study 2: Microsoft Fabric OneLake Medallion Platform - Designed a Bronze Silver Gold Fabric architecture using star schema modeling to unify ERP, CRM, files, and API sources into governed Power BI semantic models, with SQL and Python handling transformation logic across layers. Result: onboarded 3 to 4 new source systems into the platform without downtime to existing reports, and cut the time to add a new data source by an estimated 40 to 50% versus the prior report-by-report approach. Case Study 3: Databricks + PySpark Lakehouse Pipeline - Engineered a medallion layered Delta lakehouse with watermark based incremental loads, deduplication, and schema drift detection. Added a Kafka fed streaming ingestion path for near real-time event data alongside the batch lakehouse, cutting event-to-query latency from a multi-hour batch window down to under 5 seconds for time-sensitive sources. Result: reduced warehouse query runtime by an estimated 30 to 40% through partition strategy and shuffle reduction, and cut full-table reprocessing cost by avoiding unnecessary reruns on unchanged data. Core stack: Snowflake, Microsoft Fabric, Databricks, dbt, Apache Airflow, Python, SQL Also: Apache Kafka, ClickHouse, AWS (S3, Glue, Lambda), Azure Data Factory, BigQuery, PostgreSQL, PySpark, Delta Lake, Docker, Git, CI/CD Message me with how often your team catches a number that doesn't match across two reports. I'll review your current pipeline setup and tell you exactly where the break is, free.